Memory device based on memristor-diode crossbar and control CMOS logic for spiking neural network hardware

Abstract

A compact electrical circuit of a memory device based on a memristor-diode crossbar array and peripheral CMOS control logic has been developed. The peripheral logic circuit is digitally controlled and allows reading and changing the state of individual memristors. This functionality is necessary to store and transfer the synaptic states of the neural network to another neural network to avoid re-learning. Simple original electrical circuits of input and output drivers utilizing standard rectangular impulses to control memristor-diode crossbar were created. These circuits ensure the operation of the memristor matrix both as part of the hardware spiking neural network and in the modes of writing and reading the state of the memristors. Exclusion of multiple DACs and ADCs in the electrical circuits of input and output crossbar conductor drivers made it possible to significantly reduce the occupied area on the chip. On the basis of numerical modeling using the experimental characteristics of memristors the maximum size of crossbar in the developed circuit and the influence of parasitic currents on the processes of writing and reading the state of memristors are estimated. Connection of the peripheral logic circuit in the memristor leads to a limitation of the maximum size of the crossbar due to additional parasitic currents. A method of compensating the influence of parasitic currents on the process of setting memristors in a given state by varying the duration of programming pulses is proposed.